Why not just modify the Redis code to modularize the server part, and define a user friendly API? It seems like a waste to re-implement all of the data structures, since they will eventually do exactly the same thing.

This seems like a bad case of NIH syndrome.

Maybe the Redis codebase is too closely tied to the fact that the complete db has to be loaded in memory?

From the quick glance over the source code, it seems that this implementation has this assumption too. (I may be wrong, though - it is possible I've missed something.)

On the other hand, Redis has "VM" thing, which allows larger-than-memory databases by swapping less frequently used data to disk.

Yes, this was my assumption.

"It's still under a very early stage, so there is no way to test even a Beta."

Looks like an interesting idea, would love to see some no-sql storage bundled with the default rails package. Like redis, and cough memcache cough.

I would love to use Redis in non-web apps. Sqlite and Tokyo Cabinet are great, but the data structures manipulation of Redis feel much more familiar to coding, even without a data mapper.